How to get hired as a UX researcher in 2026-27

Where UX research jobs actually exist in 2026-27

Start with the thing most career advice for this role still refuses to say. Between 2023 and 2025 dedicated UX research headcount was cut harder than almost any other product function, and it was cut first because of how the job is structured: a researcher's output is an input to someone else's decision, which makes the contribution easy to underweight on a spreadsheet and easy to reassign to a product manager or designer who will do a rougher version of it. Research teams at large consumer technology companies were reduced, flattened into design organisations, or replaced with a smaller central group plus a mandate that product teams run their own studies. That is the market you are applying into. Pretending otherwise produces a job search aimed at postings that no longer exist in volume.

What did not happen is the disappearance of the work. Demand concentrated rather than vanished, and it concentrated along one clear line: how expensive it is to be wrong, and how hard it is to reach the user. If your users are consumers reachable through an in-product survey and your worst case is a mediocre quarter, a product manager with analytics and a few interviews is now considered good enough at a lot of companies. If your users are intensive care nurses, commercial underwriters, warehouse pickers, radiologists, people with motor impairments, tax filers or platform engineers, you cannot get to them with a pop-up survey, you cannot interpret what they say without domain understanding, and shipping the wrong thing costs a regulatory finding, a contract, a safety event or a year of engineering. Those are the teams still hiring researchers on purpose.

Concretely, the places where UX research survived and in some cases grew: medical devices and digital health, where usability work is part of the regulated design process rather than a nice-to-have; fintech, insurance and anything with an audit trail; enterprise and B2B software with small numbers of high-value users who sign contracts; developer tools, where the user is sceptical and the research is technical; hardware, automotive and in-car interfaces, where you cannot patch a physical control; accessibility and assistive technology; government and civic services, where the user has no alternative provider and the equity consequences of a bad flow are the point; and AI product teams that need someone who can tell them whether people actually trust, verify and recover from a model's output. Agencies and research consultancies absorbed a lot of the displaced talent and are a live hiring channel, as are staffing firms placing researchers into contract seats.

The second structural change is employment shape. A large and growing share of openings are six to twelve month contracts, vendor or agency-badged seats inside a large company, or part-time and fractional arrangements, rather than permanent headcount. This is worth looking at clearly instead of flinching. A contract seat at a company with real users gets you the shipped-decision evidence a portfolio needs, and plenty of researchers now have a career made entirely of them. The costs are real too: no equity, usually worse benefits or none, restricted access to data and meetings in some vendor arrangements, and a rolling re-interview every year. Decide deliberately which you are optimising for rather than treating contract roles as a failure state.

The third change is that the job description moved. A 2026 posting is far more likely than a 2021 one to ask for quantitative capability alongside interviewing, to ask for experience evaluating AI or model-powered features, to ask you to enable other people's research rather than only run your own, and to ask for domain familiarity by name. Read the nouns in the posting literally. A posting that says SQL, survey instrument, MaxDiff and significance is a different job from one that says contextual inquiry, field visits and service blueprint, and they are rarely well served by the same portfolio or the same first two slides.

One practical consequence of the domain shift is worth planning for before you apply: participant cost scales with how hard your population is to reach. A consumer study can be recruited from a panel in a day for a modest incentive. An hour of a cardiologist's, an underwriter's or a staff engineer's time costs a multiple of that, often needs a specialist panel or an introduction from a customer success manager, and sometimes cannot be paid for at all because the participant's employer forbids it. This is why hard-to-reach domains still employ researchers, and it is also why interviewers in those domains spend most of the design exercise on recruitment. If you are targeting clinical, financial or enterprise research, being specific about how you would actually get eight of the right people in the room is the single most convertible thing you can prepare.

• Search these titles, not just one: User Researcher, Product Researcher, Design Researcher, Mixed Methods Researcher, Quantitative UX Researcher, Research Scientist, Human Factors Engineer, Human Factors Specialist, Usability Engineer, Customer Insights, Research Operations.
• Domains still hiring on purpose: health and medical devices, fintech and insurance, enterprise and developer tools, hardware and automotive, accessibility, government and civic services, AI product teams.
• Hiring channels worth working directly: company job boards running Greenhouse, Lever or Ashby; UX research agencies and consultancies; staffing firms that place contract researchers; government job systems such as USAJOBS in the US and Civil Service Jobs in the UK; research community job boards and Slack groups.
• Signals a posting is serious about research: a named decision the role supports, participant recruitment described realistically, a research operations function mentioned, a budget for incentives, an existing researcher on the team.
• Signals a posting is a designer job with research in the title: Figma listed as a required tool, ownership of visual design deliverables, no mention of participants or recruitment, research listed last among six responsibilities.
• Before investing in an application, check for absolute blockers: citizenship or clearance requirements, mandatory on-site days for hardware and clinical work, and whether the user population can legally be recruited and paid at all.

UX researcher, product researcher, human factors engineer, market researcher: the boundaries that matter in hiring

These titles overlap enough that recruiters use them interchangeably and differ enough that applying with the wrong framing wastes the interview. A UX researcher studies how people actually use a thing in order to change what gets built: behaviour, task performance, comprehension, errors, workflow, the gap between what people say and what they do. A market researcher studies whether people will buy, which segments exist and what they will pay, and the methods overlap heavily (surveys, segmentation, conjoint) while the decisions do not. A product analyst or product data scientist works with behavioural logs at scale and can tell you precisely what happened and almost never why. Candidates who come from market research lose rounds by presenting attitudinal and preference evidence to a team that wanted task evidence, and candidates who come from purely qualitative backgrounds lose quantitative rounds by treating a survey as a way to count opinions rather than as an instrument with validity problems.

Human factors engineering is the one adjacent branch with something close to a real gate, and it is worth knowing about because it is where research is structurally protected. In medical devices, the usability work is part of the regulated design process: use-related risk analysis, identification of critical tasks whose failure can cause harm, formative studies through development, and a validation study with representative users under realistic use conditions, all documented in a usability engineering file that gets reviewed. The references to know by name are the international usability engineering standard for medical devices (IEC 62366-1), the human factors guidance published by the regulator for the market you are selling into (in the US, the FDA's human factors guidance for device submissions), and in the EU the safety and performance requirements that the same evidence feeds. Editions, amendments and submission expectations change, so name the process in an interview and check the current edition and any transition arrangements rather than quoting a version and a date from memory. The practical point for a job seeker: this work cannot be handed to a product manager with an AI summariser, because a regulator will read the file. If you have any clinical, biomedical, safety or quality background, this is the highest-leverage direction you can point a UX research job search.

Research operations is the other adjacent role that grew while research headcount shrank, and it is both a real career and a realistic way in. ResearchOps owns participant recruitment and panels, consent and privacy handling, incentives and their tax and procurement mess, tooling and licences, the research repository, the knowledge management problem of making last year's study findable, and increasingly the governance of who in the company is allowed to talk to customers and under what rules. It is operationally demanding and it sits next to every researcher in the company, which makes it a strong internal-move platform. Decide plainly which one you are applying for, because a research operations interview tests process design, vendor management and stakeholder logistics, not study design.

Finally, the internal-move reality: many working UX researchers were never hired as researchers. They were support leads, QA testers, implementation consultants, clinical specialists, sales engineers, service designers, academic researchers or teachers who were already closest to the users, started doing the work, and got the title afterwards. If you are currently inside a company with users, that path is shorter than any external application you will send this year.

• UX research answers how and why people behave, in service of what gets built. Market research answers who will buy and at what price. Product analytics answers what happened at scale. Most mixed-methods postings want the first plus enough of the other two to triangulate.
• Human factors engineering in medical devices is research with a documented process behind it: use-related risk analysis, critical tasks, formative studies, a validation study with representative users, a usability engineering file. Name the process, and verify current standard editions before quoting them.
• Research operations owns recruitment, panels, consent, incentives, tooling and the repository. Different interview, real career, strong route into research jobs.
• Service design, content design and conversation design all employ research-literate people and are worth including in a search when the research market in your domain is thin.

How UX researcher hiring actually works in 2026-27

The funnel has four layers and they test different things, so losing at each one means something different. Layer one is an applicant tracking system or a recruiter matching shallow tokens: title, domain, methods named, tools named, years. Layer two is a hiring manager spending a minute or two on your resume and, if you have one, your portfolio link, looking for a single thing: did this person's work change anything. Layer three is the live portfolio presentation, where the decision usually forms. Layer four is a set of craft and collaboration rounds that mostly confirm or sink the impression layer three created. Because the number of applicants per posting is large and the number of postings is not, every layer before the portfolio is a filter you pass rather than a place you win.

A recruiter screen is a scope, domain and logistics check. Expect a three minute background walk-through, what you owned, how many researchers on the team, what you are looking for, availability, and compensation expectations. The answer that works for the range question is the band printed on their own posting plus a question back: which level does this map to, and where in the band do offers usually land. If the posting shows no band, ask for it before you give a number. Two things sink people here, and neither is about research: a background story told as a chronology instead of a narrative with a number in it, and vagueness about whether you will accept a contract when the role is one.

The hiring manager conversation is where scoping gets tested informally. The question that carries the most weight is some version of what have you worked on that you are proud of, and the answer that works names a decision the organisation was about to make, the uncertainty that mattered, what you did, and what changed. Many researchers answer with a method and a sample size, which reads as a technician rather than a colleague. Managers are also, quietly, screening for whether you will be difficult in a specific way: researchers who treat every shipped compromise as a betrayal of the user do not get hired twice.

The portfolio presentation is the deciding round at nearly every employer that takes research seriously. Typically 45 to 60 minutes to a panel of three to six people including researchers, a designer, a product manager and sometimes an engineer, with roughly half the time expected to be questions. Two or three studies, not eight. Expect to be asked why that method, why that sample, what you would have done with two more weeks, what you got wrong, and what happened after the readout. The last question is the one that separates candidates, because it is the one that cannot be answered by somebody who left before the decision landed.

The research design exercise is usually live rather than take-home now, 45 to 60 minutes, and takes the form of a real messy prompt from their own product. Sometimes it is a critique instead: here is a study plan or a survey instrument with problems in it, tell us what is wrong. Mixed-methods and research-science postings add a quantitative round covering survey design, sampling and recruitment bias, inference and its misuse, benchmark measures such as SUS, UMUX-Lite and the single ease question, and how you would combine behavioural log analysis with interviews. Senior and staff roles add a strategy or operating round: how would you set up research for a 150 person product organisation with one researcher, what would you not do, how would you decide where your time goes.

The cross-functional panel functions as a veto, not a vote. A product manager and a designer are being asked whether they would want you in their planning meeting, and the specific risk they are checking for is a researcher who delivers findings on a timeline that does not match the roadmap, or who cannot say anything until the sample is clean. Have a concrete answer ready for how you handle a decision that has to be made on Thursday when a proper study takes three weeks, because that scenario is the job in most companies, and the honest answer (what you can learn in two days, what confidence that buys, what you would flag as still unknown, what you would watch after launch) is what they are hoping to hear.

Take-homes still exist and have changed shape. Multi-day unpaid exercises drew enough public pushback that many employers dropped them or now pay; what replaced them is a short live exercise or a request for a redacted real artefact. If you are asked for an unpaid exercise that would take more than about two hours, it is reasonable and usually safe to ask whether a live walkthrough of an existing redacted study would serve the same purpose. If you do a take-home, label your assumptions explicitly and keep it short. A twenty page research plan reads as someone who cannot prioritise under a deadline, which is the exact thing the exercise is testing.

The questions themselves repeat across employers more than candidates expect, so they can be prepared properly rather than improvised. Prepare an answer you have said out loud for each of the ones below, and prepare it as a specific story with a decision in it rather than as a principle.

• Recruiter screen, 30 min: scope, domain, availability, range. Bring a three minute story with one number in it.
• Hiring manager, 45 min: one study told as a decision that changed. Not a method list.
• Portfolio presentation, 45 to 60 min to a panel: two or three studies, half the time is questions, the decisive question is what happened after the readout.
• Research design exercise or study critique, 45 to 60 min, live: scoping, method choice, recruitment, what would change your mind.
• Quantitative and survey round for mixed-methods roles: instrument design, sampling, bias, benchmark measures, honest reading of significance.
• Cross-functional panel: a veto round about speed and collaboration, not about method depth.
• Senior and staff addition: research strategy and operating model, including what you would refuse to do.
• Questions you will actually be asked: Tell me about a study that changed a decision. Why did you choose that method over a cheaper one? How did you decide the sample? What would have changed your mind? What did you get wrong? What happened six months later? Tell me about a time your findings were ignored. How do you handle a decision that has to be made faster than a study can run? How do you decide which requests to say no to? How do you use AI tools in your analysis, and how do you keep them honest? How would you evaluate a feature whose output changes every time? What do you do when the product manager has already decided?

What UX researchers are paid, and what actually moves it

Pay for this role is unusually hard to pin down and anyone quoting you one confident national band is guessing, so here is where to look instead. The United States has no dedicated Bureau of Labor Statistics occupation code for UX research. Postings and employers map the work variously to 13-1161 Market Research Analysts and Marketing Specialists, 19-3039 Psychologists All Other, for human factors work sometimes 17-2112 Industrial Engineers, and in some employer filings 15-1255 Web and Digital Interface Designers, which is a design code and should not be read as a research rate. The BLS Occupational Employment and Wage Statistics figures for those codes are real and checkable and they understate technology-sector UX research, because they average across industries where the work is paid very differently. Use them as a floor and as evidence in a negotiation with a non-technology employer, not as your target.

The genuinely useful source is the posted range. A number of US states and cities require employers to include a pay range in job postings, which means a large share of US postings now carry a band you can read directly, including many remote roles that could be performed in a covered location. Read a dozen postings in your domain and level and you have a better picture than any survey will give you. In the UK, central government publishes Civil Service grade bands outright and the NHS publishes Agenda for Change bands, which makes public sector research roles the easiest pay research you will ever do; private sector UK postings often omit the band, so ask the recruiter in the first call. Self-reported aggregator sites are better than nothing for shape and are skewed towards large technology employers and towards people who enjoy reporting high numbers.

What actually moves the number, in rough order of effect. Industry and employer type first: a large technology company, a quantitative research-science seat, or a regulated medical device programme pays materially more than an agency, a non-profit, a university or a government department for the same craft. Level second, and levelling is where most money is won or lost in this role because research ladders are short and inconsistently mapped; a company with three research levels will slot you by scope, meaning the number of teams and the kind of decision you have influenced, not by years. Quantitative capability third: researchers who can design an instrument, write SQL, and read an experiment honestly are scarcer than researchers who can interview well, and the market prices that gap. Domain knowledge fourth, and it is larger than people expect in clinical, financial and developer tooling contexts, where a researcher who already speaks the user's language removes months of ramp.

Employment shape is the factor people most often get wrong. A contract rate quoted hourly looks like a raise and frequently is not: no employer retirement contribution, no paid leave, no equity, often no bonus, your own health coverage in the US, and gaps between contracts that are unpaid. Before accepting, convert the hourly figure into an annual number using the weeks you will actually be paid for, and subtract the benefits you are now buying yourself. Agencies pay less than in-house for comparable seniority and give you range and volume in exchange, which for someone building a portfolio is a genuine trade rather than a consolation.

On negotiation, the leverage in this role is specific. Scope evidence beats seniority claims: having influenced a decision that one named executive cared about moves a level more reliably than two more years in the title. A competing offer moves the number more than anything you say about the market. And for contract-to-hire conversations, the moment to negotiate the permanent terms is before the contract starts, not at month ten when you have already proved you are useful and have no alternative in hand.

The resume: decisions changed, not studies run

The hiring manager's scan is looking for one pattern and it is not your method coverage. It is: what decision was on the table, what did this person find, what happened differently. Almost every rejected UX research resume fails because it describes activity. Conducted 40 usability tests, ran a diary study with 12 participants, synthesised findings into actionable insights, advocated for the user across the organisation. None of that tells a reader whether anything happened, and study counts actively hurt you at senior level because they signal that volume is how you measure yourself. Write one line per decision instead, opened by the change and closed by the evidence: killed a planned redesign of the scheduling flow after finding the reported problem was a permissions issue, freeing a quarter's engineering capacity for the feature customers were actually blocked on; changed the onboarding model for clinical users after observing that nurses shared one logged-in workstation, which the original design made unsafe.

Numbers belong on this resume but they are not the numbers researchers reach for first. A sample size is a design detail, not an outcome. The numbers that land are the ones attached to the consequence: task completion before and after, the support ticket volume that fell, the error rate in a critical task, the proportion of a population affected by the problem you found with its denominator stated, engineering time saved by not building something, the renewal or contract tied to a workflow you fixed. Where you cannot share a figure because it is confidential, use a relative change or the denominator alone. An approximate number with an honest caveat beats no number, and it also demonstrates the exact habit the quantitative round is looking for.

Structure the document for the two readers it has. Top third: title, domain nouns in the user's language (intensive care nurses, commercial underwriters, Android developers, field technicians, benefit claimants), methods only as a compact line, and two or three outcome bullets from your strongest work. Then roles in reverse order with two to four outcome bullets each. A short tools and methods block near the bottom for the token match, including the ones a recruiter searches: interviews, contextual inquiry, usability testing (moderated and unmoderated), diary study, survey design, concept testing, card sorting and tree testing, benchmark measures, log analysis, SQL, participant recruitment, consent and privacy handling. Skip the summary paragraph, skip skill rating bars, skip a visual design treatment that fights the scan.

The portfolio link is part of the resume and is sometimes read before the resume. Three rules fix most portfolios. Organise by decision, not by method: a case study titled Why we did not build offline mode outperforms one titled Diary study. Lead each case with the decision and the outcome in the first paragraph, because reviewers do not scroll. And put a password-free short version first; a case study behind a request-access form or an NDA wall does not get read, and the correct move is a redacted public version with the sensitive specifics generalised (sector instead of client, direction instead of absolute figure), with the detail saved for the live presentation.

One artefact is worth building whether or not anyone asks for it: a one page readout from a real study, written the way you would send it to a product team. Decision at the top, what you found in three lines, what you recommend, what you are not sure about, what you would watch after launch. It is the thing you actually produce in the job, it takes an hour, and it answers a question no portfolio website answers, which is whether a busy reader could act on your work without you in the room.

• Do write: the decision, the finding, the change, the consequence, with a number and its denominator where you have one.
• Do not write: study counts, participant counts as achievements, synthesised insights, championed the user, drove alignment, deep empathy.
• Name the users in their own words. Clinicians, underwriters, warehouse pickers, platform engineers, tax filers. Domain nouns are what a recruiter and an applicant tracking system both match on.
• Include a kill: a study that stopped something from being built, with the cost avoided. It is the single most convincing line a researcher can put on a page.
• Include one quantitative artefact even in a qualitative career: a survey you designed, a benchmark you ran, a log analysis you partnered on. Its absence is now read as a gap.
• Name the research tooling: Dovetail, Condens, Marvin, UserTesting, Maze, Optimal Workshop, dscout, Qualtrics, User Interviews, Respondent, Prolific, Lookback, plus SQL and whichever BI tool you actually used.
• If your experience is academic, translate it. A thesis becomes a research programme with a question, a method, a constraint and a conclusion somebody acted on, and teaching becomes stakeholder communication with evidence of it.
• One page up to roughly ten years of experience, two beyond that. Portfolio link in the header, not buried at the end.

The portfolio presentation: what the panel is actually grading

This is the round that decides it, and it is misread more than any other stage in the process. Candidates prepare it as a demonstration of rigour: here is my beautifully structured study, here is my affinity map, here are my themes, here are my recommendations. The panel is grading something narrower. They want to know whether your judgement is good when conditions are bad, because that is the condition the job runs under. Every question they ask is a probe at a judgement call: why that method rather than a cheaper one, why that sample, why you believed the finding, what you did when the answer was inconvenient, and what happened after you presented.

Structure that consistently works, repeated for two or three studies in about twelve to fifteen minutes each. One: the decision, stated as a decision, with who was going to make it and when. Two: the constraint, honestly (two weeks, no budget, participants only reachable through a sales team that did not want you talking to them). Three: the design, including the thing you chose not to do and why. Four: what you found, with the two or three findings that mattered rather than all nine. Five: what the team did, including anything they did differently from your recommendation. Six: what you learned about your own method. That last beat is disproportionately persuasive and almost nobody includes it.

Bring at least one study that did not go well. A study whose sample was wrong, whose findings were ignored, whose question turned out to be the wrong question, or which concluded the feature was fine and therefore produced nothing to show. Researchers avoid this out of fear and it is the fastest credibility move available, because senior researchers on the panel have all lived it and a portfolio of three clean wins reads as either junior or curated. The requirement is that you can say what you would do differently, concretely.

Be ready for the confidentiality problem before the day. Most of your best work is under NDA. The workable approach, which panels accept routinely, is to generalise the specifics and keep the reasoning: the sector rather than the client, the direction and scale rather than the exact figure, screenshots redacted or redrawn, participant quotes stripped of identifying detail. Say once at the start what you have generalised and why. What does not work is spending the round apologising for what you cannot show, or sharing something you should not: a panel that watches you leak a former employer's data has learned something about you that no amount of craft makes up for.

Delivery details that cost people the round. Running long: half the time belongs to questions, and a candidate who fills 55 of 60 minutes has prevented the panel from doing the assessment and will be marked down for it. Reading slides aloud. Presenting a method taxonomy instead of a story. Saying the word insights more than twice. And answering what would you have done differently with nothing, which reads either as not having thought about it or as not having been there for the consequences.

If you have no professional studies to show because you are moving into the field, build two real ones rather than five exercises. Real means a genuine product with genuine users who are not your classmates, a decision somebody actually faced, and recruitment you did yourself. Workable sources: a local clinic's or charity's booking or donation flow, an open source project's documentation (its issue tracker and chat are full of reachable users who already complain in writing), a trade association or professional community where your target users talk to each other, a tool your current employer uses badly. Approach the owner, agree the question in one paragraph, run the study, deliver a one page readout, and record what they did with it. One of those with a real outcome beats a bootcamp capstone redesign of a well-known app, which panels have seen many times over and read as practice rather than evidence.

• Two or three studies, roughly twelve to fifteen minutes each, half the total time reserved for questions.
• Each study ends with what the team did, not with your recommendations.
• Include one study that went wrong and what you changed about your practice as a result.
• Generalise confidential detail up front in one sentence. Never show data you were not cleared to show.
• Expect and prepare: why this method, why this sample size, what would have changed your mind, what did you get wrong, what happened six months later.
• If you are new: two real studies with real users and a real owner who acted on them, not five classroom exercises.

The research design exercise and the quantitative round

The live design exercise is a scoping test wearing a method question's clothes. A typical prompt: our enterprise customers' administrators are not completing the user provisioning setup, support says it is confusing, the team wants to redesign the wizard next quarter, how would you investigate. A weak answer starts naming methods. A strong answer starts by establishing what decision this is for and when it has to be made, asks what is already known from logs, support tickets and past studies, and only then proposes the cheapest design that would change the decision. Say out loud the things interviewers are listening for: what you would look at before recruiting anybody, who you need in the sample and why that group rather than a broader one, how you would recruit them given this is an enterprise product, what result would point to a redesign and what result would point at documentation or permissions instead, and what you would do if you only had four days.

Name your recruitment plan concretely, because it is the part of research that most often fails in reality and the part candidates most often wave through. Who, how many, how found (existing panel, in-product intercept, customer success introductions, external recruiter, specialist panel for clinical or technical populations), what the screener asks and what it excludes, the incentive and who pays for it, and the consent and data handling. For a regulated or health context, say how you would handle personal data and whether an ethics or compliance review applies, such as an institutional review board in a US health setting or a data protection assessment where special category data is involved. Interviewers notice when a candidate knows that getting eight of the right people is harder than analysing what they say.

Expect a critique variant of the same round: here is a study plan, a screener or a survey instrument, tell us what is wrong with it. The planted problems are usually the classic ones. Leading questions. Double-barrelled items. A satisfaction scale used to answer a behavioural question. A sample recruited entirely from power users or entirely from churned accounts. A usability test script that explains the interface before asking the participant to use it. A conclusion that depends on self-reported future intent. Naming them calmly, in order of how much they threaten the conclusion, is the whole exercise.

The quantitative round is now standard for mixed-methods and research-science postings and increasingly appears in a light form elsewhere, because the market learned that a researcher who cannot read numbers gets overruled by anybody who can produce a chart. What is actually tested: survey instrument design and why question order and scale choice change the answer; sampling and the specific biases of in-product intercepts, panels and customer lists; basic inference and, more importantly, when not to use it, including the problem of a dozen comparisons on one survey; benchmark instruments such as SUS, UMUX-Lite and the single ease question, and what they can and cannot support; choice-based methods such as MaxDiff, conjoint or Kano and when a simple ranking exercise is the honest alternative; and how you pair behavioural log analysis with a qualitative study so each covers the other's weakness.

The highest-scoring answer in a quantitative round is frequently a refusal with a reason. Asked how large a sample you need for a usability study, the answer is not a number from a textbook: it is that a usability study is looking for problems rather than estimating a rate, that a handful of participants per distinct user group surfaces the frequent severe problems, and that if someone needs a number with a confidence interval around it then that is a different study with a different sample. Asked whether a two point difference in a satisfaction score between quarters is real, say what you would need to know first: sample sizes, who was sampled each time, whether the instrument changed, what else changed in the product, and whether anyone would act differently either way. Interviewers are checking whether you will let a number be over-claimed in a meeting, because the researcher is often the only person in the room who will stop it.

One more round appears at senior level and surprises people: the operating model question. One researcher, 150 people building product, unlimited demand. What do you do. The answer that works is explicit triage plus enablement plus a refusal: you take the decisions that are expensive and irreversible, you build the self-serve path for the cheap reversible ones (templates, a screener library, office hours, a reviewed usability test kit, a participant panel somebody else can use), you make past findings findable so the same question is not paid for twice, and you say out loud what you will not do. Candidates who answer by promising to serve every team are marked as people who will burn out and satisfy nobody.

• Open the exercise with the decision and its deadline, not with a method.
• Say what is already knowable from logs, tickets and prior research before proposing new research.
• Specify recruitment properly: who, how many, how reached, screener, incentive, consent, data handling.
• State the result that would change the decision, and the result that would point somewhere other than design.
• Give the four-day version as well as the three-week version. Both are part of the answer.
• In the critique variant, rank the flaws by how much they threaten the conclusion.
• Quantitative round essentials: instrument design, sampling bias, inference and its misuse, SUS, UMUX-Lite, SEQ, MaxDiff or conjoint, pairing logs with interviews.
• Be willing to say a study should not be run, and say what you would do instead.

The stakeholder round, and getting in without the title

Every panel has one round that is really about whether you will work. It usually arrives as a behavioural question: tell me about a time your findings were ignored, or about a stakeholder who disagreed with your research. The answer that fails is a story where the researcher was right, the product manager was stubborn, and the product shipped badly. Hiring managers hear that story as a prediction about you. The answer that works treats being overruled as information about how the decision was made: you find out what evidence would have been persuasive, you learn what constraint you did not know about (a contractual commitment, a launch date tied to a conference, a platform migration), and next time you arrive earlier with a smaller piece of evidence aimed at the actual decision point. Research influence is mostly timing, and candidates who understand that read as senior regardless of years.

Prepare three stories in this shape. One where you changed a decision, with the mechanism of how the evidence travelled (who you showed it to first, what format convinced them, who repeated it in a meeting you were not in). One where you were overruled and what you did next. And one where you concluded that research was not the right spend, told the team so, and spent the time somewhere else. That third story is rare and it does more for you than either of the others, because it shows you are optimising for the product rather than for research volume.

On getting in without the title, the honest ranking of routes as of late 2026. First, the internal move: you already work somewhere with users, you start doing small legitimate research in your current job, you publish the findings, and you ask for the title after you have changed two decisions. This is the shortest path by a wide margin and it is available to support leads, QA, implementation, clinical specialists, teachers, sales engineers and operations people. Second, agencies and research consultancies: they hire for craft and trainability, they expose you to many domains fast, they are well set up to teach you, and they are genuinely hiring. Third, research operations, which converts administrative and project management strength into a seat next to researchers. Fourth, government and civic digital teams, where the mission attracts people, the pay is published, the user populations are serious, and the hiring is slower but more structured; be aware that citizenship or clearance requirements can be absolute. Fifth, contract roles through staffing firms, which are plentiful and which build real evidence. Cold applications to well-known technology companies for junior research roles are last, not because they never work, but because the volume per posting makes them close to a lottery.

One thing that reliably shortcuts the lottery is being seen presenting. Researchers hire people whose thinking they have watched: a method talk at a local meetup or a conference lightning slot, a written case study in a research community, a question answered usefully in a professional Slack or forum. It is the same skill the portfolio round grades, performed in public and in advance, and it produces referrals that skip the first two layers of the funnel entirely.

If you are retraining, spend your money on the gap rather than on another general course. For a qualitative researcher the gap is almost always quantitative: a statistics or survey methodology course with real assignments, and enough SQL to pull your own behavioural data without asking anyone. For an academic the gap is speed and framing: running something useful in two weeks and presenting it in six slides to people who will not read your method section. For a designer the gap is rigour and willingness to be wrong: separating what you observed from what you concluded, and running a study whose result you did not want. A certificate from a recognised provider is acceptable vocabulary on a resume and is not what gets you the job; two real studies with named consequences is.

A last practical note on the search itself. Set alerts on the title variants rather than the one title, follow the specific agencies and consultancies in your region directly because their openings often fill before they are widely posted, and treat each domain as its own market with its own vocabulary. A researcher who learns to speak clinic workflow, claims adjudication or developer tooling can get interviews in a market where a generalist cannot, and that specialisation is the most reliable advantage available to a candidate this year.

• Have three ready stories: a decision you changed and how the evidence travelled, a time you were overruled and what you changed, and a time you said research was not the right spend.
• Routes in, best first: internal move where you already have users, agency or consultancy, research operations, government or civic digital, contract through a staffing firm, cold application to a large technology company.
• Be seen presenting: a meetup method talk, a written case study, a useful answer in a research community. It generates referrals that skip the resume screen.
• Close your actual gap: quantitative methods and SQL for qualitative researchers, speed and framing for academics, rigour and falsifiability for designers.
• Specialise by domain vocabulary. It is the cheapest competitive advantage in a thin market.

What a screen is looking for

These are the terms that a resume screen, human or automated, is matching against for this role. Use the ones that are true of you, in the words the posting uses.

• UX research
• User research
• Product research
• Design research
• Mixed methods research
• Quantitative UX research
• Qualitative research
• Generative research
• Evaluative research
• Foundational research
• Discovery research
• User interviews
• Contextual inquiry
• Field research
• Ethnographic research
• Diary study
• Usability testing
• Moderated usability testing
• Unmoderated usability testing
• Think aloud protocol
• Concept testing
• Prototype testing
• Benchmark study
• Longitudinal study
• Survey design
• Questionnaire design
• Sampling
• Screener design
• Participant recruitment
• Research operations
• ResearchOps
• Incentive management
• Informed consent
• Participant privacy
• GDPR
• Institutional review board (IRB)
• Card sorting
• Tree testing
• Information architecture research
• Heuristic evaluation
• Cognitive walkthrough
• Accessibility research
• Assistive technology testing
• WCAG
• Task analysis
• Journey mapping
• Service blueprint
• Jobs to be done
• Personas and segmentation
• Thematic analysis
• Affinity mapping
• Qualitative coding
• Triangulation
• System Usability Scale (SUS)
• UMUX-Lite
• Single Ease Question (SEQ)
• Net Promoter Score
• MaxDiff
• Conjoint analysis
• Kano model
• A/B testing
• Experiment readout
• Behavioural log analysis
• Product analytics
• SQL
• R
• Python
• Statistical significance
• Descriptive statistics
• Human factors engineering
• Usability engineering
• Use-related risk analysis
• Usability engineering file
• IEC 62366-1
• Formative and validation usability testing
• Human-computer interaction (HCI)
• Dovetail
• Condens
• Marvin
• UserTesting
• Maze
• Optimal Workshop
• dscout
• Qualtrics
• User Interviews
• Respondent
• Prolific
• Lookback
• Looker
• Figma
• Research repository
• Research enablement
• Stakeholder management
• Research readout
• Insight to decision
• AI feature evaluation
• Human evaluation of model output
• Trust and reliance research
• Annotation rubric
• AI-moderated interviews

Mistakes that cost people this job

Questions people ask